ALS

ML API

• Function

  This type of APIs is used to import sample data in dataset format, call the fit API, and output the ALS recommendation model.

• Input and output
  1. Package name: package org.apache.spark.ml.recommendation
  2. Class name: ALS
  3. Method name: fit
  4. Input: training sample data (Dataset[_])
  5. Algorithm parameters

     Algorithm Parameter

     def setRank(value: Int): ALS.this.type def setNumUserBlocks(value: Int): ALS.this.type def setNumItemBlocks(value: Int): ALS.this.type def setImplicitPrefs(value: Boolean): ALS.this.type def setAlpha(value: Double): ALS.this.type def setUserCol(value: String): ALS.this.type def setItemCol(value: String): ALS.this.type def setRatingCol(value: String): ALS.this.type def setPredictionCol(value: String): ALS.this.type def setMaxIter(value: Int): ALS.this.type def setRegParam(value: Double): ALS.this.type def setNonnegative(value: Boolean): ALS.this.type def setCheckpointInterval(value: Int): ALS.this.type def setSeed(value: Long): ALS.this.type def setIntermediateStorageLevel(value: String): ALS.this.type def setFinalStorageLevel(value: String): ALS.this.type def setColdStartStrategy(value: String): ALS.this.type def setNumBlocks(value: Int): ALS.type

  6. Added algorithm parameters

     Parameter	Description	Type
     spark.sophon.ALS.bloc kMaxRow	Row block size of a Gram matrix. Retain the default value.	Positive integer
     spark.sophon.ALS.unp ersistCycle	srcFactorRDD anti-persistence operation cycle. Retain the default value.	Positive integer

     

     An example is provided as follows:

     val als = new ALS()
         .setMaxIter(numIterations)
         .setUserCol("user")
         .setItemCol("product")
         .setRatingCol("rating")
         .setNonnegative(nonnegative)
         .setImplicitPrefs(implicitPrefs)
         .setNumItemBlocks(numItemBlocks)
         .setNumUserBlocks(numUserBlocks)
         .setRegParam(regParam)
         .setAlpha(alpha)
     val model = als.fit(ratings)

  7. Output: ALS recommendation model (ALSModel)

• Sample usage

  val model = als.fit(ratings)
  val predictions = model.transform(ratings)
  val p = predictions.select("rating", "prediction").rdd   .map { case Row(prediction: Float, label: Float) => (prediction, label) }   .map{t =>     val err = (t._1 - t._2)     err * err   }.mean() println("Mean Squared Error = " + p)

• Sample result

  Mean Squared Error = 0.9962046583156793